Pose Forecasting in Industrial Human-Robot Collaboration (ECCV2022)

Utils for "Pose Forecasting in Industrial Human-Robot Collaboration" (project page) paper accepted at ECCV'22.

The code for the forecasting is available here. In this repository there is a torch dataloader for the inspection of the dataset, with utils for visualization using Open3D.

NOTE: the suggested version for open3d is 0.15.2, as the API keeps changing and the older versions may differ a lot.

Requirements:

Python 3.x (suggested 3.8+)

• Numpy
• PyTorch
• open3d>=0.15.2
• trimesh

Dataset

The dataset is available here and presents both 3D poses (for human and robot) and the RGB video frames.

Run the visualization

The current code visualize the 3D poses only.

First, download the dataset (see section Dataset), and prepare the repo with the following structure:

.
├── data
│   └── chico
│       ├── poses
│       │   ├── S00
│       │   │   ├── hammer.pkl
│       │   │   ├── lift.pkl
│       │   │   ├── place-hp.pkl
|       |   |   ...
│       │   ├── S01
│       │   │   ├── hammer.pkl
│       │   │   ├── lift.pkl
│       │   │   ├── place-hp.pkl
|       |   |   ...
|       |   ...
│       └── rgb
│           ├── S00
│           │   ├── 00_03.mp4
│           │   ├── 00_06.mp4
│           │   └── 00_12.mp4
│           ├── S01
│           │   ├── 00_03.mp4
│           │   ├── 00_06.mp4
│           │   └── 00_12.mp4
|           ...
├── open3d_wrapper.py
├── chico_dataset.py
└── show_poses.py

Note that the RGB is optional, and not included in the visualization tool.

The code to run is show_poses.py.